Rubin

in English: Nvidia Rubin (Vera Rubin platform)

advancedspreading

Rubin is Nvidia's chip generation following Blackwell: a combination of processors and network elements for data centers, on which large AI models are trained and run.

In plain terms

When you ask a chatbot a question, the answer is not generated on your phone but in a remote data center where thousands of specialized chips work simultaneously. Rubin is Nvidia's new generation of such chips, the successor to the earlier generation called Blackwell.

Imagine it as a new turbine type in a power plant: at home, you only notice that there is electricity, but how much you get and at what price depends on the turbine. Rubin is the "turbine" of AI services: it determines how fast and how cheaply you get an answer.

Rubin is not a single chip but a coordinated set: a graphics accelerator (Rubin GPU), a central processor (Vera CPU), and network elements that move data between them. Nvidia designs these together and sells them as fully assembled, rack-sized systems.

Nvidia promises that it does the same amount of work at much lower cost and power use than its predecessor. This matters especially for AI agents that work independently over long periods, because they require far more computation than a simple question and answer.

How it works

Most of the work is done by the Rubin GPU: this chip performs countless tiny calculations in parallel, exactly as AI models require.
Alongside it, the Vera CPU acts as the conductor: it prepares the data, assigns tasks, and performs general-purpose operations that do not require the GPU.
Very fast memory is placed alongside the chips so that the model's billions of numerical values are immediately available, and waiting does not consume time.
Many dozens of such chips go into one rack – for example, the Vera Rubin NVL72 deployment includes 72 Rubin GPUs and 36 Vera CPUs – connected by a fast internal network, so they behave externally as a single giant computer.
The cabinets are cooled with liquid and installed by the thousands in data centers – these are often called 'AI factories'.

An everyday example

You read in the news that a cloud provider is building a new data center worth billions of dollars with 'Vera Rubin systems'. This means that the next version of the chatbot or coding agent you use will likely run on such machines.

Who is affected

bank customer, investorThe stock price of Nvidia and its suppliers is strongly influenced by whether the Rubin arrives on time and in the expected quantity. → Money

entrepreneur, executiveThe efficiency of the new chip generation will eventually appear in the price and available capacity of AI services.

developerGreater computing capacity makes larger models and longer-running agents more affordable. → Code

industrial worker, engineerThe power demand and liquid cooling of Rubin cabinets pose new tasks for data center designers and operators. → Industry

A common misconception

MythRubin is a new graphics card you will buy in stores for your computer.

In realityThe Rubin data center platform: sold in entire racks to cloud providers and large enterprises. You encounter it indirectly through AI services.

Limits and risks

Performance and cost figures provided by the manufacturer are promises; the actual result depends on the task and the software.
A Rubin cabinet consumes a vast amount of electricity, so in many places the bottleneck is not the chip but the electrical grid and cooling.
Export restrictions apply to the most advanced Nvidia chips (primarily toward China), so they are not available in every country or to every buyer.
Anyone building on the Nvidia platform is strongly tied to a single manufacturer's tools and software, making switching expensive.

Similar, but not the same

Blackwell – Blackwell is Nvidia's previous data center generation; Rubin is its successor.
Vera CPU – Vera is the platform's central processor, and Rubin is the graphics accelerator. The full platform name is Vera Rubin, but common usage often refers to the entire generation simply as Rubin.
Vera C. Rubin Observatory – This is a Chilean observatory named after the same astronomer, but it has no connection to the Nvidia product.

Related concepts

Broader

MI-chip, AI infrastructure, Data center

Narrower

Rubin GPU, Vera CPU, Rubin Ultra

On the edge

Blackwell, GPU, Nvidia, AI factory, computing capacity

A brief history

Nvidia announced Rubin as the generation after Blackwell in June 2024 at the Computex exhibition in Taiwan, and presented the platform in detail in January 2026 at CES. It is named after Vera Rubin, an American astronomer whose measurements became important evidence for the existence of dark matter.

Where things stand

Nvidia announced in January 2026 that production of the platform had started and promised the first Rubin systems in the second half of 2026, initially for major cloud providers. The rollout is gradual: in most operating data centers, Blackwell is still working for now.

Status: 2026 November

In the news

Appears in 6 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Nvidia's Kyber racks could require 600 kW of continuous powerNov 5, 2026 · Ipar
Schneider and Nvidia take digital twin to live 1.2 GW siteOct 26, 2026 · Ipar
Foxconn AI factory runs on Nvidia cluster; Nvidia and Foxconn target eight-week AI factory deploymentOct 26, 2026 · Ipar
Nvidia unveils DSX POD, a container-sized AI factory it says can be deployed in 14 daysOct 21, 2026 · Ipar
India's AI moves: sovereign stack, 9,000 Vera Rubin chips, $1.5 billion Porsche–TCS dealAug 24, 2026 · Agent Age
SpaceX unveils Starmind satellite network to bring AI compute to space with Nvidia Rubin GPUsAug 9, 2026 · Ipar

Also known as

Nvidia Rubin · Vera Rubin · Vera Rubin platform · Rubin GPU · Rubin-platform · Rubin chip · Rubin Ultra · Vera Rubin NVL144 · Vera Rubin NVL72

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.